pci: Fix pci_device_iommu_address_space() bus propagation
[qemu/kevin.git] / cpu-exec.c
blob713540fc8fe40f14c627e093f682f9c604f4ea22
1 /*
2 * emulator main execution loop
4 * Copyright (c) 2003-2005 Fabrice Bellard
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
19 #include "config.h"
20 #include "cpu.h"
21 #include "trace.h"
22 #include "disas/disas.h"
23 #include "tcg.h"
24 #include "qemu/atomic.h"
25 #include "sysemu/qtest.h"
26 #include "qemu/timer.h"
27 #include "exec/address-spaces.h"
28 #include "exec/memory-internal.h"
29 #include "qemu/rcu.h"
30 #include "exec/tb-hash.h"
32 /* -icount align implementation. */
34 typedef struct SyncClocks {
35 int64_t diff_clk;
36 int64_t last_cpu_icount;
37 int64_t realtime_clock;
38 } SyncClocks;
40 #if !defined(CONFIG_USER_ONLY)
41 /* Allow the guest to have a max 3ms advance.
42 * The difference between the 2 clocks could therefore
43 * oscillate around 0.
45 #define VM_CLOCK_ADVANCE 3000000
46 #define THRESHOLD_REDUCE 1.5
47 #define MAX_DELAY_PRINT_RATE 2000000000LL
48 #define MAX_NB_PRINTS 100
50 static void align_clocks(SyncClocks *sc, const CPUState *cpu)
52 int64_t cpu_icount;
54 if (!icount_align_option) {
55 return;
58 cpu_icount = cpu->icount_extra + cpu->icount_decr.u16.low;
59 sc->diff_clk += cpu_icount_to_ns(sc->last_cpu_icount - cpu_icount);
60 sc->last_cpu_icount = cpu_icount;
62 if (sc->diff_clk > VM_CLOCK_ADVANCE) {
63 #ifndef _WIN32
64 struct timespec sleep_delay, rem_delay;
65 sleep_delay.tv_sec = sc->diff_clk / 1000000000LL;
66 sleep_delay.tv_nsec = sc->diff_clk % 1000000000LL;
67 if (nanosleep(&sleep_delay, &rem_delay) < 0) {
68 sc->diff_clk = rem_delay.tv_sec * 1000000000LL + rem_delay.tv_nsec;
69 } else {
70 sc->diff_clk = 0;
72 #else
73 Sleep(sc->diff_clk / SCALE_MS);
74 sc->diff_clk = 0;
75 #endif
79 static void print_delay(const SyncClocks *sc)
81 static float threshold_delay;
82 static int64_t last_realtime_clock;
83 static int nb_prints;
85 if (icount_align_option &&
86 sc->realtime_clock - last_realtime_clock >= MAX_DELAY_PRINT_RATE &&
87 nb_prints < MAX_NB_PRINTS) {
88 if ((-sc->diff_clk / (float)1000000000LL > threshold_delay) ||
89 (-sc->diff_clk / (float)1000000000LL <
90 (threshold_delay - THRESHOLD_REDUCE))) {
91 threshold_delay = (-sc->diff_clk / 1000000000LL) + 1;
92 printf("Warning: The guest is now late by %.1f to %.1f seconds\n",
93 threshold_delay - 1,
94 threshold_delay);
95 nb_prints++;
96 last_realtime_clock = sc->realtime_clock;
101 static void init_delay_params(SyncClocks *sc,
102 const CPUState *cpu)
104 if (!icount_align_option) {
105 return;
107 sc->realtime_clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL_RT);
108 sc->diff_clk = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) - sc->realtime_clock;
109 sc->last_cpu_icount = cpu->icount_extra + cpu->icount_decr.u16.low;
110 if (sc->diff_clk < max_delay) {
111 max_delay = sc->diff_clk;
113 if (sc->diff_clk > max_advance) {
114 max_advance = sc->diff_clk;
117 /* Print every 2s max if the guest is late. We limit the number
118 of printed messages to NB_PRINT_MAX(currently 100) */
119 print_delay(sc);
121 #else
122 static void align_clocks(SyncClocks *sc, const CPUState *cpu)
126 static void init_delay_params(SyncClocks *sc, const CPUState *cpu)
129 #endif /* CONFIG USER ONLY */
131 void cpu_loop_exit(CPUState *cpu)
133 cpu->current_tb = NULL;
134 siglongjmp(cpu->jmp_env, 1);
137 /* exit the current TB from a signal handler. The host registers are
138 restored in a state compatible with the CPU emulator
140 #if defined(CONFIG_SOFTMMU)
141 void cpu_resume_from_signal(CPUState *cpu, void *puc)
143 /* XXX: restore cpu registers saved in host registers */
145 cpu->exception_index = -1;
146 siglongjmp(cpu->jmp_env, 1);
149 void cpu_reload_memory_map(CPUState *cpu)
151 AddressSpaceDispatch *d;
153 if (qemu_in_vcpu_thread()) {
154 /* Do not let the guest prolong the critical section as much as it
155 * as it desires.
157 * Currently, this is prevented by the I/O thread's periodinc kicking
158 * of the VCPU thread (iothread_requesting_mutex, qemu_cpu_kick_thread)
159 * but this will go away once TCG's execution moves out of the global
160 * mutex.
162 * This pair matches cpu_exec's rcu_read_lock()/rcu_read_unlock(), which
163 * only protects cpu->as->dispatch. Since we reload it below, we can
164 * split the critical section.
166 rcu_read_unlock();
167 rcu_read_lock();
170 /* The CPU and TLB are protected by the iothread lock. */
171 d = atomic_rcu_read(&cpu->as->dispatch);
172 cpu->memory_dispatch = d;
173 tlb_flush(cpu, 1);
175 #endif
177 /* Execute a TB, and fix up the CPU state afterwards if necessary */
178 static inline tcg_target_ulong cpu_tb_exec(CPUState *cpu, uint8_t *tb_ptr)
180 CPUArchState *env = cpu->env_ptr;
181 uintptr_t next_tb;
183 #if defined(DEBUG_DISAS)
184 if (qemu_loglevel_mask(CPU_LOG_TB_CPU)) {
185 #if defined(TARGET_I386)
186 log_cpu_state(cpu, CPU_DUMP_CCOP);
187 #elif defined(TARGET_M68K)
188 /* ??? Should not modify env state for dumping. */
189 cpu_m68k_flush_flags(env, env->cc_op);
190 env->cc_op = CC_OP_FLAGS;
191 env->sr = (env->sr & 0xffe0) | env->cc_dest | (env->cc_x << 4);
192 log_cpu_state(cpu, 0);
193 #else
194 log_cpu_state(cpu, 0);
195 #endif
197 #endif /* DEBUG_DISAS */
199 cpu->can_do_io = !use_icount;
200 next_tb = tcg_qemu_tb_exec(env, tb_ptr);
201 cpu->can_do_io = 1;
202 trace_exec_tb_exit((void *) (next_tb & ~TB_EXIT_MASK),
203 next_tb & TB_EXIT_MASK);
205 if ((next_tb & TB_EXIT_MASK) > TB_EXIT_IDX1) {
206 /* We didn't start executing this TB (eg because the instruction
207 * counter hit zero); we must restore the guest PC to the address
208 * of the start of the TB.
210 CPUClass *cc = CPU_GET_CLASS(cpu);
211 TranslationBlock *tb = (TranslationBlock *)(next_tb & ~TB_EXIT_MASK);
212 if (cc->synchronize_from_tb) {
213 cc->synchronize_from_tb(cpu, tb);
214 } else {
215 assert(cc->set_pc);
216 cc->set_pc(cpu, tb->pc);
219 if ((next_tb & TB_EXIT_MASK) == TB_EXIT_REQUESTED) {
220 /* We were asked to stop executing TBs (probably a pending
221 * interrupt. We've now stopped, so clear the flag.
223 cpu->tcg_exit_req = 0;
225 return next_tb;
228 /* Execute the code without caching the generated code. An interpreter
229 could be used if available. */
230 static void cpu_exec_nocache(CPUState *cpu, int max_cycles,
231 TranslationBlock *orig_tb)
233 TranslationBlock *tb;
235 /* Should never happen.
236 We only end up here when an existing TB is too long. */
237 if (max_cycles > CF_COUNT_MASK)
238 max_cycles = CF_COUNT_MASK;
240 tb = tb_gen_code(cpu, orig_tb->pc, orig_tb->cs_base, orig_tb->flags,
241 max_cycles | CF_NOCACHE);
242 tb->orig_tb = tcg_ctx.tb_ctx.tb_invalidated_flag ? NULL : orig_tb;
243 cpu->current_tb = tb;
244 /* execute the generated code */
245 trace_exec_tb_nocache(tb, tb->pc);
246 cpu_tb_exec(cpu, tb->tc_ptr);
247 cpu->current_tb = NULL;
248 tb_phys_invalidate(tb, -1);
249 tb_free(tb);
252 static TranslationBlock *tb_find_slow(CPUState *cpu,
253 target_ulong pc,
254 target_ulong cs_base,
255 uint64_t flags)
257 CPUArchState *env = (CPUArchState *)cpu->env_ptr;
258 TranslationBlock *tb, **ptb1;
259 unsigned int h;
260 tb_page_addr_t phys_pc, phys_page1;
261 target_ulong virt_page2;
263 tcg_ctx.tb_ctx.tb_invalidated_flag = 0;
265 /* find translated block using physical mappings */
266 phys_pc = get_page_addr_code(env, pc);
267 phys_page1 = phys_pc & TARGET_PAGE_MASK;
268 h = tb_phys_hash_func(phys_pc);
269 ptb1 = &tcg_ctx.tb_ctx.tb_phys_hash[h];
270 for(;;) {
271 tb = *ptb1;
272 if (!tb)
273 goto not_found;
274 if (tb->pc == pc &&
275 tb->page_addr[0] == phys_page1 &&
276 tb->cs_base == cs_base &&
277 tb->flags == flags) {
278 /* check next page if needed */
279 if (tb->page_addr[1] != -1) {
280 tb_page_addr_t phys_page2;
282 virt_page2 = (pc & TARGET_PAGE_MASK) +
283 TARGET_PAGE_SIZE;
284 phys_page2 = get_page_addr_code(env, virt_page2);
285 if (tb->page_addr[1] == phys_page2)
286 goto found;
287 } else {
288 goto found;
291 ptb1 = &tb->phys_hash_next;
293 not_found:
294 /* if no translated code available, then translate it now */
295 tb = tb_gen_code(cpu, pc, cs_base, flags, 0);
297 found:
298 /* Move the last found TB to the head of the list */
299 if (likely(*ptb1)) {
300 *ptb1 = tb->phys_hash_next;
301 tb->phys_hash_next = tcg_ctx.tb_ctx.tb_phys_hash[h];
302 tcg_ctx.tb_ctx.tb_phys_hash[h] = tb;
304 /* we add the TB in the virtual pc hash table */
305 cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)] = tb;
306 return tb;
309 static inline TranslationBlock *tb_find_fast(CPUState *cpu)
311 CPUArchState *env = (CPUArchState *)cpu->env_ptr;
312 TranslationBlock *tb;
313 target_ulong cs_base, pc;
314 int flags;
316 /* we record a subset of the CPU state. It will
317 always be the same before a given translated block
318 is executed. */
319 cpu_get_tb_cpu_state(env, &pc, &cs_base, &flags);
320 tb = cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)];
321 if (unlikely(!tb || tb->pc != pc || tb->cs_base != cs_base ||
322 tb->flags != flags)) {
323 tb = tb_find_slow(cpu, pc, cs_base, flags);
325 return tb;
328 static void cpu_handle_debug_exception(CPUState *cpu)
330 CPUClass *cc = CPU_GET_CLASS(cpu);
331 CPUWatchpoint *wp;
333 if (!cpu->watchpoint_hit) {
334 QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) {
335 wp->flags &= ~BP_WATCHPOINT_HIT;
339 cc->debug_excp_handler(cpu);
342 /* main execution loop */
344 volatile sig_atomic_t exit_request;
346 int cpu_exec(CPUState *cpu)
348 CPUClass *cc = CPU_GET_CLASS(cpu);
349 #ifdef TARGET_I386
350 X86CPU *x86_cpu = X86_CPU(cpu);
351 CPUArchState *env = &x86_cpu->env;
352 #endif
353 int ret, interrupt_request;
354 TranslationBlock *tb;
355 uint8_t *tc_ptr;
356 uintptr_t next_tb;
357 SyncClocks sc;
359 /* This must be volatile so it is not trashed by longjmp() */
360 volatile bool have_tb_lock = false;
362 if (cpu->halted) {
363 if (!cpu_has_work(cpu)) {
364 return EXCP_HALTED;
367 cpu->halted = 0;
370 current_cpu = cpu;
372 /* As long as current_cpu is null, up to the assignment just above,
373 * requests by other threads to exit the execution loop are expected to
374 * be issued using the exit_request global. We must make sure that our
375 * evaluation of the global value is performed past the current_cpu
376 * value transition point, which requires a memory barrier as well as
377 * an instruction scheduling constraint on modern architectures. */
378 smp_mb();
380 rcu_read_lock();
382 if (unlikely(exit_request)) {
383 cpu->exit_request = 1;
386 cc->cpu_exec_enter(cpu);
388 /* Calculate difference between guest clock and host clock.
389 * This delay includes the delay of the last cycle, so
390 * what we have to do is sleep until it is 0. As for the
391 * advance/delay we gain here, we try to fix it next time.
393 init_delay_params(&sc, cpu);
395 /* prepare setjmp context for exception handling */
396 for(;;) {
397 if (sigsetjmp(cpu->jmp_env, 0) == 0) {
398 /* if an exception is pending, we execute it here */
399 if (cpu->exception_index >= 0) {
400 if (cpu->exception_index >= EXCP_INTERRUPT) {
401 /* exit request from the cpu execution loop */
402 ret = cpu->exception_index;
403 if (ret == EXCP_DEBUG) {
404 cpu_handle_debug_exception(cpu);
406 cpu->exception_index = -1;
407 break;
408 } else {
409 #if defined(CONFIG_USER_ONLY)
410 /* if user mode only, we simulate a fake exception
411 which will be handled outside the cpu execution
412 loop */
413 #if defined(TARGET_I386)
414 cc->do_interrupt(cpu);
415 #endif
416 ret = cpu->exception_index;
417 cpu->exception_index = -1;
418 break;
419 #else
420 cc->do_interrupt(cpu);
421 cpu->exception_index = -1;
422 #endif
426 next_tb = 0; /* force lookup of first TB */
427 for(;;) {
428 interrupt_request = cpu->interrupt_request;
429 if (unlikely(interrupt_request)) {
430 if (unlikely(cpu->singlestep_enabled & SSTEP_NOIRQ)) {
431 /* Mask out external interrupts for this step. */
432 interrupt_request &= ~CPU_INTERRUPT_SSTEP_MASK;
434 if (interrupt_request & CPU_INTERRUPT_DEBUG) {
435 cpu->interrupt_request &= ~CPU_INTERRUPT_DEBUG;
436 cpu->exception_index = EXCP_DEBUG;
437 cpu_loop_exit(cpu);
439 if (interrupt_request & CPU_INTERRUPT_HALT) {
440 cpu->interrupt_request &= ~CPU_INTERRUPT_HALT;
441 cpu->halted = 1;
442 cpu->exception_index = EXCP_HLT;
443 cpu_loop_exit(cpu);
445 #if defined(TARGET_I386)
446 if (interrupt_request & CPU_INTERRUPT_INIT) {
447 cpu_svm_check_intercept_param(env, SVM_EXIT_INIT, 0);
448 do_cpu_init(x86_cpu);
449 cpu->exception_index = EXCP_HALTED;
450 cpu_loop_exit(cpu);
452 #else
453 if (interrupt_request & CPU_INTERRUPT_RESET) {
454 cpu_reset(cpu);
456 #endif
457 /* The target hook has 3 exit conditions:
458 False when the interrupt isn't processed,
459 True when it is, and we should restart on a new TB,
460 and via longjmp via cpu_loop_exit. */
461 if (cc->cpu_exec_interrupt(cpu, interrupt_request)) {
462 next_tb = 0;
464 /* Don't use the cached interrupt_request value,
465 do_interrupt may have updated the EXITTB flag. */
466 if (cpu->interrupt_request & CPU_INTERRUPT_EXITTB) {
467 cpu->interrupt_request &= ~CPU_INTERRUPT_EXITTB;
468 /* ensure that no TB jump will be modified as
469 the program flow was changed */
470 next_tb = 0;
473 if (unlikely(cpu->exit_request)) {
474 cpu->exit_request = 0;
475 cpu->exception_index = EXCP_INTERRUPT;
476 cpu_loop_exit(cpu);
478 spin_lock(&tcg_ctx.tb_ctx.tb_lock);
479 have_tb_lock = true;
480 tb = tb_find_fast(cpu);
481 /* Note: we do it here to avoid a gcc bug on Mac OS X when
482 doing it in tb_find_slow */
483 if (tcg_ctx.tb_ctx.tb_invalidated_flag) {
484 /* as some TB could have been invalidated because
485 of memory exceptions while generating the code, we
486 must recompute the hash index here */
487 next_tb = 0;
488 tcg_ctx.tb_ctx.tb_invalidated_flag = 0;
490 if (qemu_loglevel_mask(CPU_LOG_EXEC)) {
491 qemu_log("Trace %p [" TARGET_FMT_lx "] %s\n",
492 tb->tc_ptr, tb->pc, lookup_symbol(tb->pc));
494 /* see if we can patch the calling TB. When the TB
495 spans two pages, we cannot safely do a direct
496 jump. */
497 if (next_tb != 0 && tb->page_addr[1] == -1) {
498 tb_add_jump((TranslationBlock *)(next_tb & ~TB_EXIT_MASK),
499 next_tb & TB_EXIT_MASK, tb);
501 have_tb_lock = false;
502 spin_unlock(&tcg_ctx.tb_ctx.tb_lock);
504 /* cpu_interrupt might be called while translating the
505 TB, but before it is linked into a potentially
506 infinite loop and becomes env->current_tb. Avoid
507 starting execution if there is a pending interrupt. */
508 cpu->current_tb = tb;
509 barrier();
510 if (likely(!cpu->exit_request)) {
511 trace_exec_tb(tb, tb->pc);
512 tc_ptr = tb->tc_ptr;
513 /* execute the generated code */
514 next_tb = cpu_tb_exec(cpu, tc_ptr);
515 switch (next_tb & TB_EXIT_MASK) {
516 case TB_EXIT_REQUESTED:
517 /* Something asked us to stop executing
518 * chained TBs; just continue round the main
519 * loop. Whatever requested the exit will also
520 * have set something else (eg exit_request or
521 * interrupt_request) which we will handle
522 * next time around the loop.
524 next_tb = 0;
525 break;
526 case TB_EXIT_ICOUNT_EXPIRED:
528 /* Instruction counter expired. */
529 int insns_left = cpu->icount_decr.u32;
530 if (cpu->icount_extra && insns_left >= 0) {
531 /* Refill decrementer and continue execution. */
532 cpu->icount_extra += insns_left;
533 insns_left = MIN(0xffff, cpu->icount_extra);
534 cpu->icount_extra -= insns_left;
535 cpu->icount_decr.u16.low = insns_left;
536 } else {
537 if (insns_left > 0) {
538 /* Execute remaining instructions. */
539 tb = (TranslationBlock *)(next_tb & ~TB_EXIT_MASK);
540 cpu_exec_nocache(cpu, insns_left, tb);
541 align_clocks(&sc, cpu);
543 cpu->exception_index = EXCP_INTERRUPT;
544 next_tb = 0;
545 cpu_loop_exit(cpu);
547 break;
549 default:
550 break;
553 cpu->current_tb = NULL;
554 /* Try to align the host and virtual clocks
555 if the guest is in advance */
556 align_clocks(&sc, cpu);
557 /* reset soft MMU for next block (it can currently
558 only be set by a memory fault) */
559 } /* for(;;) */
560 } else {
561 /* Reload env after longjmp - the compiler may have smashed all
562 * local variables as longjmp is marked 'noreturn'. */
563 cpu = current_cpu;
564 cc = CPU_GET_CLASS(cpu);
565 cpu->can_do_io = 1;
566 #ifdef TARGET_I386
567 x86_cpu = X86_CPU(cpu);
568 env = &x86_cpu->env;
569 #endif
570 if (have_tb_lock) {
571 spin_unlock(&tcg_ctx.tb_ctx.tb_lock);
572 have_tb_lock = false;
575 } /* for(;;) */
577 cc->cpu_exec_exit(cpu);
578 rcu_read_unlock();
580 /* fail safe : never use current_cpu outside cpu_exec() */
581 current_cpu = NULL;
582 return ret;